A statistical form reading system

In this paper, we introduce a system able to read statistical forms. The system is composed of a personal computer, a scanner and a form reading software. For a form image, the system first extracts the form lines, then locates the individual rectangle fields of the form, obtains the relation between them and sends image segment of each field to character recognition module. In character recognition, we use the AQ15 machine learning system to generate the classifier. The result can be stored in text or in database, with or without form lines. When one kind of form is first input (we refer to this form as unlearned), the attribute of each field column is determined by man-machine interaction, then the structure and the public fields of the form is recorded. When a learned form is input, the structure of the form is compared with the original structure. If no conflict, the system recognizes non-public fields consisting of numerals or printed Chinese characters, while the other fields can be input friendly.<>